2009/10/17 Mathias Fröhlich <mathias.froehl...@gmx.net>
> As always, tell the exact problem.
>
> Depending on your problem and requirements the solution ranges from a few
> lines of code to - 'better use an implementation that already exists is
> tested
> and is numerically stable'.
>
Hi Mathias,
I will be receiving a sequence of 2d data points in "real time". I will
start by assuming a linear relationship/fit which I know in advance is a
reasonable assumption. I would like to find a way to incrementally compute
a simple straight line least squares fit of the data I have received so
far. I know incremental approaches exist. Isaias sent me a simple
approach, but this maintains sums of all the data received so far and as
Alex pointed out, that will be subject to increasing round off errors as the
data accumulates (this code could be receiving hundreds of data points per
second over the course of hours, days, even weeks.)
So yes, a numerically stable approach is important. I suspect the code will
just be a few lines, so if I can find an approach that is laid out
algorithmically or in terms of some sort of pseudo-code, I'm pretty sure I
can create and test my own implementation.
Maybe I'm only imagining that such a thing exists, I googled for quite a
while yesterday on a variety of search terms that are directly or loosely
related and wasn't able to turn up what I was hoping to find. (Thus my cry
for help) :-)
A method that "forgets" the oldest data and weights newer data more heavily
might also be interesting (versus an approach that sums up the entire
history of the data ... although that would be ok too.) I'm happy to start
simple and get fancier later on if I need to.
Thanks,
Curt.
--
Curtis Olson: http://baron.flightgear.org/~curt/
------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Flightgear-devel mailing list
Flightgear-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/flightgear-devel
